Multicriterion evaluation of animal welfare : example of on-farm dairy cows

There is an increasing societal concern for the welfare of farm animals. The European project Welfare Quality® aims at developing a standard for the assessment of and information about animal welfare. Welfare is composed of several dimensions (health, behaviour, absence of stress...). Hence, its overall evaluation is based on a set of complementary measures. The objective of this PhD thesis is to develop a model for the overall assessment of the welfare of dairy cows, which respects the multidimensionality of the welfare concept. Our model follows a multicriterion approach. It follows a sequential evaluation process where several criteria are constructed and aggregated into an overall assessment. After identifying the constraints imposed on the construction of an evaluation model of overall welfare, we define a set of 4 criteria (Good feeding, Good housing, Good health and Appropriate behaviour ) and 12 subcriteria. To construct the subcriteria from the 44 measures observed on farms, we propose to assess them on a common utility scale [0,100]. Raw data are thus aggregated using methods adapted to the number of measures, their nature and their relative importance (weighted sums associated to a utility function, decision trees...). Subcriteria involved in the same criterion are the aggregated using a method which allows to consider that subcriteria are more or less important, while limiting compensations between subcriteria (Choquet integral). Finally, the aggregation of criteria into an overall assessment is based on a comparison to predefined profiles delimiting four categories of welfare (corresponding to an ordered sorting procedure). At each stage, the evaluation model is parameterized on the basis of experts' opinions. The experts are scientists or potential users of the future evaluation system). The evaluation system that will be produced by Welfare Quality® (in 2009) can be used for several purposes : a compulsory European labelling system, a welfare-friendly labelling system, advice to farmers or research (with the capacity of the model to explain the results being adequate for these two later purposes).